The Effects of Market Reform on Trading Costs of Public Investors: Evidence from the London Stock Exchange. Narayan Y Naik and Pradeep K Yadav*

Transcription

1 The Effects of Market Reform on Trading Costs of Public Investors: Evidence from the London Stock Exchange by Narayan Y Naik and Pradeep K Yadav* First draft: November 1998 This draft: June 1999 * Narayan Naik is from the London Business School, Sussex Place, Regent's Park, London NW1 4SA, United Kingdom (tel , fax , nnaik@lbs.ac.uk). Pradeep Yadav is from the Department of Accounting and Finance, University of Strathclyde, 100 Cathedral Street, Glasgow G4 0LN, United Kingdom (tel , fax , p.k.yadav@strath.ac.uk). We would like to thank Donald Campbell and Etleva Skenderi for excellent research assistance; thank Nicola Sawford, Matthew Leighton, Steven Wells and Graham Hart of the London Stock Exchange for providing the data and for several useful discussions; and thank Mike Barclay, John Board, Richard Brealey, Paul Dawson, Gordon Gemmill, Marc Lipson, Ron Masulis, Paul Schultz, Erik Sirri, S. Vishwanathan, seminar participants at Groupe ESSEC and participants of the NASDAQ Notre Dame Microstructure conference, SIRIF Trading Costs Conference, and NYSE-Paris Bourse Global Markets Conference for useful comments. Both authors are grateful for research grant R from the UK Economic and Social Research Council (ESRC). The authors are also grateful for financial support from the European Commission's TMR program grant (network reference ERBFMRXCT ) and thank the Scottish Institute for Research in Investment and Finance (SIRIF) for intra structural support. The authors remain responsible for all errors.

2 The Effects of Market Reform on Trading Costs of Public Investors: Evidence from the London Stock Exchange Abstract In October 1997, the London Stock Exchange removed the obligation of dealers to quote firm two-way prices for FTSE 100 index stocks, and allowed the public to compete directly with dealers in these stocks through the submission of limit orders. This article examines the effects of these market reforms on the trading costs of public investors, the targeted beneficiary of the reforms, and documents several interesting results. First, the duly signed average effective half-spread of public investors has decreased much more than the corresponding decrease in the absolute effective half-spread documented by Barclay et. al. (1998) for NASDAQ. This is because a sub-set of public investors trade through limit orders, and thereby earn the spread rather than pay it. Second, consistent with the change from obligatory to voluntary market making, there is a significant increase in the positioning revenue earned by dealers from a change in the price of a stock while they are carrying the stock in their inventory. As a result, the overall gain of public investors in terms of the realised half-spread is not significantly different from zero. Third, the cross-subsidisation across trade sizes has disappeared, leading to a significant decline in the average execution costs of small public trades and an increase for large public trades. Fourth, the market reforms have caused negative externalities for stocks not going through the new trading system. Finally, in the absence of the price stabilisation provided earlier by dealers, the inside half-spread has increased very sharply in the first hour of trading a finding which highlights the need for special opening procedures for electronic order books.

3 The Effects of Market Reform on Trading Costs of Public Investors: Evidence from the London Stock Exchange There has been a long-standing debate among market practitioners, regulators and academics, on the relative merits of quote driven dealer markets versus order driven auction markets. Until recently, the London Stock Exchange (hereafter LSE) and the NASDAQ were the two main equity markets organised as pure competitive dealer markets. In 1997, both these markets underwent significant market reforms: NASDAQ on January 20, 1997 and LSE on October 20, In both cases, an active dealer market continued to exist but the public began to compete directly with dealers through the submission of binding limit orders. However, the LSE reform represented a significantly greater move from a dealer-type market to an order driven market. Unlike NASDAQ, the dealers on LSE were no longer obliged to quote firm two-way prices as they did earlier. Their participation became entirely voluntary, and their quotes were no longer available to investors through publicly available price-display systems. Barclay et. al. (1999) document the effects of these market reforms on the trading costs of NASDAQ stocks and Bessembinder (1998) compares relative trading costs on NASDAQ and NYSE after the reforms. In this paper, we investigate the effects of the LSE market reforms on the trading costs of public investors. We examine not only the stocks which underwent market reforms, but also a sample of stocks which were not included in the first phase of reform 1. 1 This serves two purposes. First, it provides us an estimate of inter-temporal variation in trading costs caused by other market wide factors, such as changes in volatility, trading volume etc. Second, it enables examination of externalities on stocks not undergoing reform. It was not possible for Barclay et. al. (1999) to do this since the reforms on NASDAQ were quickly extended to all stocks. In contrast, on the LSE the extension of reform to stocks not included in the first phase has been very slow. 1

4 This (London based) study makes several major contributions to the existing (NASDAQ based) research in this area. First, our data distinguishes the trades done by dealers as principals from the trades done by individual or institutional public investors, or agents acting on their behalf. This allows us to analyse trading costs from the frame of reference of a public investor, the targeted beneficiary of these market reforms. It also enables us to examine the extent to which public investors, as against market intermediaries, act as liquidity suppliers rather than liquidity demanders, and thereby earn the spread instead of paying the spread, and the impact this has on the net trading costs of public investors as a whole 2. Existing US studies have analysed effective half-spreads earned (paid) by liquidity suppliers (demanders) without distinguishing whether the liquidity supplier (demander) was a market intermediary (dealer or broker or floor trader) or a public investor. We analyse trading costs faced by the end-users of the trading system, i.e. the individual or institutional public investors, without confounding our analysis by the costs faced by intermediaries. We find that the duly signed average effective half-spreads of public traders (as a group) has fallen after the market reform much more massively than the decrease in absolute effective half-spreads documented for NASDAQ by Barclay et. al. (1999). This is primarily because a large sub-set of public investors chose to post orders on the limit order book, rather than demand immediate execution, thereby earning the spread rather than paying it. Second, we examine the impact of the change from obligatory market making to voluntary market making on the positioning revenues of dealers, i.e. revenues arising from a change in the price of a stock while the intermediary is carrying the stock in her inventory. Before the market reforms on the LSE, dealers were required to always stand 2 We are not aware of any work that reports the trading costs of public investors as a whole. Keim and Madhavan (1995, 1996, 1998), Chan and Lakonishok (1993, 1997), Edwards and Wagner (1993), and Perold and Sirri (1993) use data on institutional trades and data from specific market intermediary firms and institutions to document the trading costs of a subset of public investors. We are also not aware of any work that examines 2

5 ready to buy and sell the stocks in which they made a market. These dealers could be expected to lose money, on average, in trading with more informed investors (Glosten and Milgram, 1985), and consistent with this expectation, Hansch et. al. (1999) and Sofianos (1995) document negative overall positioning revenues of obligatory market makers on LSE and NYSE respectively. However, after the reforms on LSE, dealers can be selective about the stocks and trades in which they choose to participate. Arguably, conditional on information at the time of the trade, their expected profits from any inventory they choose to hold should be greater than zero for risk averse traders 3. In this context, unlike the existing studies examining the impact of market reforms, we also investigate the impact of the reform on positioning revenues of market intermediaries. We find that positioning revenues are significantly negative before the reforms, and continue to be significantly negative after the reforms for stocks that did not undergo trading system reform. However, consistent with the expectation of positive inventory holding profits for voluntary market makers, we find that average positioning revenues of dealers are positive for post-reform trades, whether executed through the voluntary dealer network or the electronic limit order book. In view of the significant increase in positioning revenues, there is no significant change in the overall trading revenues of dealers even though the revenues they earn from the effective half-spread have significantly decreased. In other words, the overall gain of public investors, as measured by the average change in the realised half-spread, is not significantly different from zero, even though the average signed effective half-spread paid by public investors has reduced significantly because they have earned the spread through posting of limit orders. specifically the effects of trading system reforms on the trading costs of any subset of public investors. 3 Bernhardt and Hughson (1997) show that since a limit order book is split against incoming market orders, equilibrium limit order schedules yield positive expected profits to agents posting them. 3

6 Third, we find that, on average, market participants, both intermediaries and public investors, who supply liquidity through posting of limit orders earn the spread but lose on the post-trade price change; and market participants who demand liquidity through market orders pay the spread but earn on the post-trade price change. This implies that, consistent with the results of Harris and Hasbrouck (1996) for NYSE stocks, and Berkman (1996) for EOE options, limit orders are subject to adverse selection risk. More importantly, we find no difference with respect to post-trade price changes between limit (or market) orders posted by market intermediaries and limit (or market) orders posted by public investors. In both cases we find that about sixty percent of the post-trade price change arises within 15 minutes of a trade, reflecting imminently stale limit orders being picked off, and the other forty percent arises over the rest of day. This suggests that aggressor market participants picking limit orders have superior short-term information. Fourth, we quantify the effective cross-subsidisation of dealership services taking place across different size-categories of trades, and hence analyse changes in this crosssubsidisation caused by the market reform. As highlighted by Domowitz and Steil (1999), in the pure dealership London market prior to the market reform the profits of dealers in small trades were effectively subsidising trading costs of large trades. However, we find after the introduction of the limit order book, this cross-subsidisation has disappeared, leading to a significant decrease in the average execution costs of small public trades and an increase in the average execution costs of large public trades. Fifth, we document negative externalities in terms of significantly wider inside half-spreads during the first hour of trading, significantly higher effective half-spreads, significantly higher realised half-spreads and significantly lower price improvement (relative to best 4

7 quotes) for stocks that did not undergo trading system reform. This suggests that the market reforms have led to changes in the pattern of cross- subsidisation and the flow of information across different categories of stocks. Finally, we document very significant qualitative changes in the intra-day variation of the inside half-spread. In the quote driven pure dealership market, the inside half-spread displays no significant intra-day variation. This is arguably because of the price stabilisation provided by dealers. However, in the order driven system after the reform, in the absence of this price stabilisation, the inside half-spread is sharply higher in the first hour of trading, declines significantly over the first 60 minutes, and then steadily until the close. Our results clearly highlight the need for special order matching procedures at the open in order driven markets, since market participants appear to be reluctant to expose their limit orders immediately after market opening while the price uncertainty after a long non-trading interval is being resolved. The rest of this paper is organised as follows. Section I outlines the market reforms and explains the motivation behind them. Section II describes the data used in our empirical work and outlines the empirical research design. Section III examines intra-day variation in inside half-spreads. Section IV presents the results of the trading system change on signed effective half-spreads defined from the frame of reference of an individual or institutional public investor. Section V similarly reports the impact on realised half-spreads and the adverse information component of realised half-spreads. Section VI analyses the positioning and trading revenues of market intermediaries over different post-trade intervals when public investors supply liquidity through limit orders or demand liquidity through market orders. Finally, our concluding remarks appear in Section VII. 5

8 I Brief History & Description of Market Reforms A. Brief History The London equity market had always been a pure dealership quote driven market. All trades were executed through competing dealers who were obliged to make firm two-way prices in the stocks they wished to make a market in. In 1997, the LSE introduced an order driven system for the most liquid stocks while retaining the dealer network in parallel. This reform was the biggest structural change in the history of the LSE 4. Several developments led to the introduction of this major market reform: First, the nineties saw a significant increase in the competitive pressures on the LSE from various directions. Many European exchanges introduced new trading systems to capture the market share in European equities from the LSE, though none of these represented structural changes like LSE s change from a dealership to an order driven market 5. Second, alternative electronic networks such as Tradepoint established themselves in the UK, and the UK regulatory authorities removed the earlier restrictions on market makers that had prevented them from quoting, on other electronic networks, prices better than those they were quoting on LSE. Thus, these electronic networks had the potential of bringing buyers and sellers together well within the LSE spread. Finally, under Article 15.4 of EU Investing Services Directive, automated order matching systems located anywhere within Europe became potential competitors of LSE 6. 4 A set of important changes made by the LSE in October 1986, known as the Big Bang, made transaction prices and quotes more transparent, allowed dual capacity trading, abolished fixed commissions and liberalised the entry of foreign firms. However, the Big Bang retained the pure dealership nature of the market. While quote display systems were computerised, the trade execution function was not automated. 5 Many European Exchanges introduced changes in their trading systems to make themselves more competitive: TSA by Amsterdam Stock Exchange in 1994; SWX by the Swiss Exchange in 1995; XETRA by Deutsche Borse AG in 1997; and NSC by the Paris Bourse in 1996 (see Demarchi and Foucault, 1998). 6 This article allowed European markets to enrol remote members in other EU countries without securing permission from the regulatory authorities in that country. 6

9 Market participants and researchers started comparing costs of trading the same stocks across different exchanges 7. Although LSE served well the needs of large domestic institutional investors through its high depth 8, it was widely perceived as a market with high trading costs for small retail investors and this high cost was attributed to the dealership nature of the market 9. Since LSE and NASDAQ were the two main dealer markets in equities, the controversies created by the collusion allegations on NASDAQ (Christie and Schultz (1994), Christie, Harris and Schultz (1994)) added to the popular perception that LSE market makers were profiting unduly from the dealership system. And amid all these developments, LSE was perceived to be sitting still rather than doing something about the changed market conditions in the nineties. B. Nature of Market Reforms In October 1997, the exchange introduced a fundamental shift in the nature of the market by replacing the quote-driven manual trade execution system to an order-driven electronic trade execution system interacting with a network of dealers. In the new order-driven system, buyers and sellers could post limit orders or pick limit orders electronically (through their broker or a member firm), and they could also trade with dealers functioning as voluntary liquidity suppliers. Since order-driven systems seem to work better for liquid stocks, the change was introduced initially for the hundred highest market capitalisation stocks which 7 See Pagano and Roell (1990) and De Jong et. al., (1995) for comparison of quoted and inside spreads across LSE and European exchanges. Also see Werner and Kleidon (1996) for the analysis of UK and US trading of British cross-listed stocks. 8 Large institutional US investors were apparently not happy with the lack of transparency in the London market making system since it appears they were paying higher trading costs relative to domestic institutional traders. This is clear, for example, from the April 1999 newsletter of the Plexus Group, and the background note circulated by Global Investor for their July 1999 roundtable discussion in London on institutional trading. 9 The perception of high trading costs in the old dealer type London market is not entirely correct. Although the quoted half-spreads and inside half-spreads on LSE had been relatively high, the overall value-weighted average of effective half-spreads and the overall value-weighted average of realised half-spreads have been documented to be very low (Hansch et al (1999), Naik and Yadav (1999)). However, these effective and realised half-spreads differed significantly across different trade sizes, being large in magnitude for small trades, small in magnitude for medium-to-large trades, and increasing again in magnitude for very large trades. 7

10 were part of the FTSE-100 index. This coverage has been extended (albeit very slowly) to other stocks as new stocks have joined the FTSE-100 after quarterly market capitalisation reviews, and in some other special cases. Relative to the market reforms on the NASDAQ, the reforms on LSE represented a significantly greater move from a quote-driven dealership market to an order-driven auction market. Unlike the NASDAQ, dealers were no longer obliged to quote firm two way prices as they did earlier, and their quotes were no longer available to investors through publicly available price-display systems. Dealers participation became entirely voluntary, and the functions of price formation and liquidity creation were left to collective market forces. Yet, unlike other order driven markets like Paris and Frankfurt, LSE did not introduce special procedures for the market open. However, an active dealer market did continue to formally exist with limited obligations to interact with the order book. C. Expected Benefits Compared to the old dealer-market, the LSE expected the order book to result in lower trading costs of public investors because of several reasons. First, the new system would give public investors more choice, thus opening up the potential for a range of different trading strategies 10. The ability to post limit orders also improved the bargaining power of public investors while negotiating with dealer firms. Second, the public investors were now able to observe the entire order book and have real-time knowledge of prices and quantities of trades executed on the order book. Third, order handling costs were expected to be lower due to automated order execution. Thus, the wider choice, greater transparency and lower order 10 The limit order book gives the public investor an important tool to influence the trading costs. Either the public trader can pay the inside half-spread and guarantee execution immediately, or she can try to earn that effective half-spread by being patient and posting a limit order on the electronic order book. Execution of a limit order is not guaranteed, but if it does execute, she can earn the spread instead of paying it. 8

11 processing costs were expected to significantly improve the competitiveness of the LSE 11. D. After the Reform Since October 20, 1997 trades in SETS stocks can be done either through the electronic order book or through the dealer network offering dealership services on a voluntary basis. About 20% to 30% of public trades go through SETS. Most of the public trading through SETS is in medium- sized trades in the 10,000 to one NMS size range 12. The rest are mainly executed through dealers. In general, the average trade size on the electronic order book increases significantly with firm size. The proportion of inter-firm principal-to-principal trades that take place through the electronic order book is significantly greater than the proportion of public customer trades going through SETS. This is partially because the inter-dealer trades that used to be executed anonymously through the (since discontinued) Inter-Dealer-Broker (IDB) order-matching network in the old pure dealer market now go through SETS 13. Less than 2% of the trades on SETS are executed directly between the agents of two public customers. Market intermediaries trading on their own account are the counter-parties in more than 98% of the public trades executed through SETS. Hence, dealers continue to play a major role in execution of trades though they do so voluntarily and are not obliged to do so. 11 There is strong interest among regulators, practitioners, exchange members and market participants in analysing whether these expected benefits have materialized (see London Financial News, 5 October 1998). 12 NMS is the normal market size of a particular stock. It is defined as being equal to approximately 2.5% of the average total daily trading volume in that stock over a reference three-month period. Thus, the trade execution difficulty for medium to large trades becomes more comparable across stocks when trade size is expressed in multiples of NMS. 13 See e.g. Reiss and Werner (1998, 1999) for a detailed analysis of inter-dealer trading on the LSE under the old dealership system. 9

12 II Data and Methodology In order to examine the effect of market reform on the trading costs of public investors, we investigate three periods, each of three months duration. We use data from May 1998 to July 1998 (hereafter labelled 1998 ) to examine trading costs after the reform. We use two control periods before the market reform: August 1994 to October 1994 (hereafter labelled 1994 ) and February 1996 to April 1996 (hereafter labelled 1996 ). 14 Changes in trading costs in non-sets stocks during these periods act as a proxy for longer term market-wide movements over the three sample periods. For each of these three periods, we analyse comprehensive time stamped trades, quotes and limit orders data provided by the London Stock Exchange. The data for 1994 and 1996 is for the pure dealership market prior to the market reform. The data for 1998 also includes data from the SETS electronic order book. In particular, it includes details of entry and execution of all orders on the electronic order book and execution of all trades off the order book in the SETS stocks. Our trades data indicates the dealing capacity of the Stock Exchange member firm reporting the trade, i.e., whether the firm acted as principal or as an agent in the particular transaction and whether the firm bought or sold in that trade. Hence, we do not have to use an arbitrary rule to decide if the trade was a public buy or a public sell. More importantly, this enables us to determine whether it was a Stock Exchange member firm or a public customer who was effectively supplying liquidity (and earning spreads) in the trade. 14 Our choice of periods is determined entirely by availability of the data from the LSE. Transactions data from the pre-reform quote driven pure dealership market on the LSE has been used by various researchers to explore several interesting issues relating to trading costs. Reiss and Werner (1995) provide early evidence on the subject; Tonks and Snell (1995) analyse the components of the bid-ask spread; Gemmill (1996) and Board and Sutcliffe (1995) examine the impact of different transparency rules on trading costs; Hansch et. al. (1999) examine the effect of preferencing, internalization and best execution on trading costs; and Naik and Yadav 10

13 We analyse the impact of market reforms on four measures of trading costs: inside half-spread, signed effective half-spread which when aggregated across all public trades determines the spread revenue of market intermediaries, realised half-spread which when aggregated across all public trades determines the trading revenue of market intermediaries, and positioning half-spread which when aggregated across all public trades determines the positioning revenue of market intermediaries and represents the part of spread revenues they lose due to adverse selection. Let i p t be the transaction price for the i th trade executed at time t. Since transactions can potentially change the contemporaneous best limit prices in an order driven market, we measure best quotes or limit prices one second before the execution of the trade. We define a t as the lowest ask quote or limit price to sell one second before time t, and b t as the highest bid quote or limit price to buy one second before time t. Let m t be the mid-price of best quotes or limit orders at time t, = ( a + b ) / 2. Let the subscript T denote the particular m t t t post-trade reference time used to estimate the true value of the stock.: in this paper T is taken as t+15, or t+30, or t+60, or the end of the day. We define and analyse the following from spread variables: Inside Half-Spread IS t as IS t ( at bt ) / 2 = m t (1999) investigate differences in the quality of execution offered by different dealers and different brokers. 11

14 Signed Effective Half-Spread i ES t for the i th trade as ES i t i ( pt mt ) = m t for a public buy ES i t i ( mt pt ) = m t for a public sell Realised Half-Spread i RS t for the i th trade as RS i t i ( pt mt ) = m t for a public buy RS i t i ( mt pt ) = m t for a public sell Positioning Half-Spread i PS t for the i th trade as PS i t = RS ES. Thus, i t i t PS i t ( mt mt ) = m t for a public buy PS i t ( mt mt ) = m t for a public sell Our analyses are based on time-weighted averaging of inside half-spreads IS t, and valueweighted averaging over the relevant trade size categories of the other three spread variables ES, RS and i t i t i PS t. Our data covers all stocks, but as SETS has been introduced and planned only for the top 350 stocks, we confine our analyses to the 245 stocks (among the FTSE350 stocks in 1998) for which we have data for each of the three periods above. These include 76 SETS stocks which we use to examine the impact of market reforms, and 169 non-sets stocks which we use to estimate the extent of inter-temporal market-wide changes over the years. We measure each of the four spread variables in basis points (i.e., one-hundredths of a percentage point) and However, the focus of this paper is totally different from earlier research in this area. 12

15 examine differences from 1994 to 1998, and from 1996 to 1998, for SETS stocks as well as for non-sets stocks. The statistical significance of pair-wise differences in means over these sample periods is determined using standard t-tests. We sort and analyse inside half-spreads based on the eight different hourly trading intervals within the day. We sort and analyse signed effective half-spreads, realised half-spreads and positioning half-spreads based on six different trade size bands defined in terms of the "normal-market size" (NMS) of the stock as follows: below 4000; from 4000 to 10000; from to 25000; from to one NMS; from one NMS to three NMS; and from three NMS to eight NMS. Since the SETS order book system is virtually unused for trades above eight NMS, we do not analyse trades above eight NMS in size. Table I reports the distribution of trading volume across different trade size bands in each of the three sample periods. Panel A presents the value of shares traded and Panel B the corresponding number of shares traded. Both panels lead to similar inferences. Overall, there has been substantial growth in trading volume from 1994 to 1996, but not from 1996 to For non-sets stocks, the cross-sectional distribution of trades across different trade size bands has not changed significantly from 1994 to 1996 or from 1996 to However, for SETS stocks, while the cross-sectional distribution of trades across different trade size bands has not changed significantly from 1994 to 1996, it has changed significantly from 1996 to 1998 (p-value of Chi-square statistic << 0.001). This is because, relative to 1996 or 1994, a substantially greater proportion of trades have taken place in the middle trade size ranges in In particular, 82% by value (and 52% by number) of all trades executed through the order book are in the to one NMS size range. This suggests a greater propensity on the part of market participants to split large trades and work the order through. 13

16 III Inside Half-Spread For the dealer market in 1994 and 1996, and for non-sets stocks in 1998, the inside halfspread is calculated from the best bid quote and the best ask quote. For SETS stocks in 1998, the inside half-spread is calculated from the best limit orders to buy and sell. For each of the 245 sample stocks, we take 5-minute snapshots of the inside half-spread for each day in each of the three sample periods: 1994, 1996 and We then calculate average values of the inside half-spreads across different stocks and different 5-minute periods of the day. Table II reports these average inside half-spreads, separately for SETS stocks and non-sets stocks (i.e. stocks not traded on the new order book system). The table also reports the change in inside half-spreads between 1994 & 1998, and 1996 & 1998, and gives the statistical significance of pair-wise differences in means over these sample periods using standard t-tests. Figure I and Figure II show the intra-day variation in average inside half-spreads of SETS and non-sets stocks respectively over the trading day in each of the years 1994, 1996 and Before we examine the changes in SETS stocks, it is useful to examine the changes in non- SETS stocks (which did not undergo market reform) to get an idea of the general market trend over these periods. The average inside half-spread of non-sets stocks declined from 71 basis points in 1994 to 60 basis points in 1996 to 57 basis points in 1998, which indicates a general decline in inside half-spreads over time. The decline from 1994 to 1998 is large in magnitude and statistically highly significant (p-value «0.0001) within each hourly interval of the day. The decline from 1996 to 1998 is small in magnitude and not significant for any hourly interval within the day but is statistically significant overall (p-value<0.01). 14

17 Qualitatively, the changes are not different for different hours of the day. However, the average inside half-spread over the first hour is significantly higher than the average inside half-spread over the rest of the day. This difference is about 3 basis points in 1994 and in 1996 and about 5 basis points in Interestingly, the increase in this difference from 1994 to 1998 and from 1996 to 1998 is statistically significant (p-value<0.05). In sharp contrast, Figure 1 shows the very significant qualitative and quantitative changes in the inside half-spread of SETS stocks (which underwent the market reform) from 1994 to 1998, or from 1996 to The overall average inside half-spread declined from 32 basis points in 1994 to 27 basis points in 1996 and 27 basis points in Even though there is little change from 1996 to 1998 in the overall average figure, there is a massive increase in average inside half-spread in the first hour of trading after the open, and a significant decrease in the average inside half-spread for the rest of the trading day. In 1994 and 1996, there is virtually no intra-day variation in the average inside half-spread across different hours of the day. However, after the introduction of the electronic order book, the average inside half-spread in the first five minutes of trading in 1998 is more than 90 basis points - more than three times the average inside half-spread over rest of the trading day. The inside half-spread on the order book declines monotonically up to the close of trading on the day, but the average inside half-spread over the first hour of trading is still more than double the average inside half-spread over rest of the trading day, and the difference is statistically highly significant (p-value «0.0001). As a result, there is a large highly significant (p-value «0.0001) increase in average inside half-spreads in the first hour 15 The data for inside half-spread that we use for 1998 sample period is from May 1, to July 19, This is because the exchange changed the opening on 20 July 1998 from 8.30 am to 9.00 am. The results for the last 10 days of July are virtually identical to the rest of the 1998 period when the hours are measured from the new opening time. 15

18 of trading, from 28 basis points in 1996 (or 34 basis points in 1994) to 54 basis points in There is also a significant (p-value <0.01) increase in average inside half-spreads in the second hour of trading as we go from 1996 to However, since the inside halfspread on the order book declines steadily over the day, there is a significant decrease in average inside half-spreads (p-value <0.05) for the remaining hours of the day as we go from 1996 (or 1994) to If we exclude the first hour of the trading day, the average inside half-spread in 1998 is about 15% less than in 1996 and about 30% less than in In both cases the decrease is statistically highly significant (p-value «0.01), a finding totally consistent with that of Barclay et. al. (1999) for the NASDAQ stocks. The figures in Table II are time weighted average inside half-spreads. We also calculate the average inside half-spreads one second before the execution of every public trade 16, and find that the average of these trade-time inside half-spreads is about 10% to 40% lower (at different times of the day) than the figures we report for average time weighted inside halfspreads. This suggests that public traders do not usually trade at times when the inside halfspread is abnormally high. However, the intra-day variation is qualitatively very similar. To summarise, we would like to emphasise two implications of the above findings. First, our results show that dealers fulfil a useful price stabilisation function. There exists extensive evidence that open-to-open returns are more volatile than close-to-close returns, and that this is related to the greater degree of price uncertainty at the open in view of the long non-trading interval that precedes the open (Stoll and Whaley (1990) and Amihud and Mendelson (1987; 1991)). The resolution of this price uncertainty takes place after commencement of trading. Limit orders effectively provide free options which other market participants can pick off, 16 In view of the way the inside half-spread is defined for the order driven market, we calculate the average inside half-spread one second before the execution of the trade as the trade itself can potentially change the 16

19 and our results are consistent with market participants in the order driven system being justifiably reluctant to expose their limit orders during the period for which this price uncertainty is being resolved after the open. On the other hand, in the pure dealer market system preceding the market reform, and for the stocks not subject to the reform, dealers are fulfilling a price stabilisation function and absorbing the shocks associated with the greater price uncertainty immediately after the market opening. After the reform, dealers on LSE are not expected to fulfil that stabilisation function leading to large intra-day variation in the inside half-spread. Second, the introduction of the electronic limit order book for the heavily traded stocks in the market has generated negative externalities for the less liquid stocks for which no limit order book has been introduced. Even though there is has been a (significant) 5% decline in the average inside half-spread from 1996 to 1998, the intra-day variation in this spread, as proxied by the difference between the first trading hour and other trading hours, has almost doubled. The increase in the difference is significant (p-value<0.05). Clearly, dealers face greater risks in low market capitalisation low trading volume non-sets stocks relative to high market capitalisation high trading volume SETS stocks, and hence it is not surprising that there exists significant intra-day variation in inside half-spreads even in 1994 and But the level of price stabilisation provided by market makers in 1998 to non-sets stocks has significantly reduced, even though these stocks have not gone through reforms. We believe that this is driven by the large increase in the inside half-spreads of SETS stocks in the first hour. Since the true value could be anywhere within the best bid and the best ask, the greater uncertainty introduced by the large inside half-spreads of FTSE-100 index stocks estimated inside half-spread. 17

20 makes it difficult for participants trading non-sets stocks to infer where the main market is, or which way it is going. Clearly, the size of the trades that could be executed at the inside half-spread in the pure dealer market in 1994 and 1996 differs from the size of the trades that could be executed at the inside half-spread on the limit order book in In 1994 and 1996, each market maker quote was typically firm up to the "normal-market size" (NMS) of the particular stock, and it was possible to "hit" several competing market makers in the stock at the same time. However, it is difficult to estimate the true depth of the SETS stocks in 1998 because the limit orders on the book and the dealers trading off the limit order book together constitute the depth in the SETS stocks. The depth of the limit order book as seen on the screen is only about one NMS. However, this is a misleading estimate of the true depth because of the existence of substantially greater, albeit non-quantifiable, hidden depth in the dealer network. Therefore, the schedule of effective half-spreads paid by different sized trades is a more meaningful indicator of true depth than the explicitly visible schedule of limit orders posted on the screen. IV Signed Effective Half-Spreads for a Public Investor While the inside half-spread is based on posted prices at which an investor could trade, the effective half-spread is based on the prices at which an investor actually trades. The inside half-spread can be different from the effective half-spread because trades can take place both within and outside the inside half-spread. The best ask and bid quotes and limit prices are valid only up to a specific trade size. Large buy (sell) trades may have to take place above (below) the ask (bid) price in order to attract liquidity suppliers to take on the additional risk involved. However, in less transparent regimes, large trades can at times get significantly 18

21 better prices than the ask or bid prices as dealers implicitly "purchase" the information contained in the order flow (Naik et. al., 1999). Dealers can also offer attractive transaction prices in order to restore their temporary inventory imbalances to normal levels (Hansch et. al., 1998). For either of these reasons, dealers may even "reach across the market" to buy (sell) above (below) the mid-price (Sofianos, 1995). Thus, under different market conditions, the effective half-spreads can be very different for different sized trades and for different counter-parties, a phenomenon well documented in Reiss and Werner (1996). The effective half-spread of executed trades is calculated from the duly signed difference between the transaction price and the mid-price. We sign the effective half-spread from the perspective of a public customer. It is positive when the public customer pays the spread to a dealer supplying immediacy in a transaction. It is negative when the public customer earns the spread, for example, when a limit order posted by the customer is picked off by a dealer, or when a dealer buys (sells) blocks of stock above (below) the mid-price for inventory control or information reasons. In such transactions, the dealer is effectively seeking immediacy and paying the spread. In contrast to our use of duly-signed effective half-spreads, major US studies (eg Barclay, 1997; Barclay et. al., 1999; Bessembinder, 1997, 1998; Christie and Huang, 1994; Christie et. al., 1994; and Huang and Stoll, 1996) are based on the absolute value of the difference between the trade price and the mid-price. This is because the data typically available in the US does not flag trade direction. This absolute value of effective half-spread generically measures the execution costs of liquidity demanders on the assumption that a liquidity demander always trades above the mid-price for a buy and below the mid-price for a sell and does not distinguish between market intermediaries and public investors. By using signed 19

22 effective half-spreads, we are able to distinguish the times when public investors act as liquidity suppliers from times when they act as liquidity demanders. We calculate the signed effective half-spread for each trade in each of the three sample periods. We divide the 1998 trades in SETS stocks into two parts: those executed automatically through the order book, and those executed off the order book. For the purpose of our investigation, we consider only public trades in which a public customer (individual or institution), or the agent of a public customer, trades with a member firm trading as principal. We do not include the tiny fraction of public trades (less than 2%) in which a public customer (or her agent) trades directly with another public customer (or her agent) 17. For the sub-set of direct customer-to-customer trades, the average of the signed effective spread would clearly be zero 18. We sort these trade-by-trade effective half-spreads into 48 cells based on the eight different hourly trading intervals within the day and the six different trade size bands indicated in section II. For each of the 245 sample stocks separately, we calculate the average signed effective half-spread in each of these (8 6), i.e. 48 classifications based on eight trading intervals and six trade size intervals. Our final inferences are based on the , i.e average signed effective half-spreads so calculated and conducting statistical tests on the pair wise differences from 1994 to 1996 to This ensures that the inferences we report are not unduly influenced by clustering of trades (by number or by market value) in any particular trade size category or in any particular trading hour. 17 We do not also include inter-dealer trades since our analysis is confined to public investors who are not a counter-party in these trades. 18 Because of this, the signed effective spread measure is not necessarily appropriate for markets where a substantial proportion of trades are customer-to-customer trades executed through agents. 20

23 Table III reports the average signed effective half-spreads for different sized trades, separately for SETS stocks and non-sets stocks, for each of the years 1994, 1996 and It also reports the change in average signed effective half-spreads between 1994 & 1998, and 1996 & Table III has at least three interesting features. First, the average signed effective half-spread for trades executed through the order book in 1998 is either indistinguishable from zero or significantly negative, for all trade size bands. This is because public traders have often supplied liquidity to the market through posting of limit orders, thereby earning the spread rather than paying it. This has significantly reduced the average spread paid by public traders as a group. It is interesting in this context to examine the proportion of cases in which public investors are supplying liquidity through limit orders rather than demanding liquidity through market orders. We find that for small trades, i.e. less than in size, and for large trades, i.e. more than three NMS in size, public investors supply liquidity through limit orders in about two-thirds of all electronic order book trades and demand liquidity through market orders in the remaining one-third. For medium sized trades, i.e. between and three NMS in size, public investors supply liquidity through limit orders in about half the electronic order book trades and demand liquidity through market orders in the remaining half. Second, the variation of average signed effective half-spread across different trade size bands is qualitatively similar in 1994 and 1996 for both SETS and non-sets stocks but qualitatively different in 1998 particularly for SETS stocks. In 1994 and 1996, both for SETS and non-sets stocks, effective half-spreads are large and significantly positive for relatively smaller trade sizes up to about in trade size. However, for larger trade sizes, effective 21

24 half-spreads become statistically indistinguishable from zero or significantly negative. This suggests that the level of prices being offered by dealers in these trade size bands are often based on their inventory control considerations or because they are effectively paying for the information contained in the (potentially informed) order flow. However, in 1998, for SETS stocks traded through the dealer network, average signed effective half-spreads are either indistinguishable from zero or are significantly negative for relatively smaller sized trades and significantly positive only for relatively large sized trades. In the earlier pure dealer system, there was cross subsidisation taking place: the revenues of dealers from small trades were subsidising their losses from large trades (Domowitz and Steil, 1999). Our results indicate that the competitive pressures of SETS for smaller sized trades have eliminated this cross-subsidisation. Third, the effective half-spreads charged by dealers for non-sets stocks have increased significantly (p-values «0.01) across the board for all trade sizes. The average value-weighted effective spread has more than doubled from about 15 basis points in 1994 and 1996 to about 33 basis points in 1998, even though, as we saw in Section III, the average inside half-spread for these stocks in 1998 has declined by about 5% from 1996 and 20% from The majority of trades in the London quote driven dealer market have typically taken place within the spread, with dealers typically offering different levels of price improvement (with reference to quoted prices) for different trades dependent on the perceived information content of the trade, the level of their inventory and the nature of the client 19. Our results indicate that the range of price improvement being offered by dealers in different types of trades in non-sets stocks has decreased substantially after the introduction of the electronic order book in SETS stocks. This suggests that the dealers are making up for the profits lost 19 See e.g. Reiss and Werner (1995), Hansch et. al. (1999). 22

25 due to increased competition on SETS by changing higher effective half-spreads for non- SETS stocks. If this indeed is the case, then this potentially reflects a change in the pattern of cross-subsidisation in dealer services across different categories of stocks. V. Positioning Half-spreads and Realised Half-spreads The effective half-spread paid by public investors is the visible component of the execution cost at the time of the trade. It is a measure of execution cost on the assumption that the contemporaneous mid-price is the true value of the stock. On the other hand, the realised half-spread is a measure of execution cost on the assumption that the post-trade mid-price is the true value of the stock. The positioning half-spread is the difference between the two and reflects the post-trade price change. If a subset of public investors have superior information, then they will benefit, on average, from the post-trade price change, and hence the realised half-spread paid by them will be lower than the effective half-spread by the extent of the positioning half-spread. The trading revenue of all market intermediaries over any particular period represents the total cost paid by public investors as a group for trading during that period. It is equal to the duly signed and value weighted sum of realised half-spreads across all public trades in that period. This trading revenue can be considered as the sum of two components the spread revenue and the positioning revenue 20 which similarly equal the duly signed and value weighted sum across all public trades of effective half-spreads and positioning half-spreads respectively. The positioning revenue represents the gain or loss made by the intermediary due to carrying of inventory. If the public investor sells the stock to the intermediary before a price fall or buys before a price rise, potentially because of superior information, then positioning 20 See Sofianos (1995) and Hansch et. al. (1999). 23